HASSL: Hierarchy-Aware Self-Supervised Learning Framework for Single Cell Microscopy

2026-07-05

AI Summary

This work addresses the limitation of existing self-supervised learning methods in single-cell microscopy imaging, which often neglect hierarchical structures and consequently obscure fine-grained morphological features under coarse-grained factors. To overcome this, the authors propose the first self-supervised learning framework that explicitly models hierarchical organization by integrating segmentation-based teacher distillation with a hierarchical contrastive loss derived from HDBSCAN. This approach effectively disentangles coarse- and fine-grained features, yielding biologically meaningful cellular subclusters. Evaluated across 20 datasets encompassing 2.3 million cells, the method achieves an average top-K accuracy improvement of 2.8%, a 6.3% gain in top-9 retrieval performance at the deepest hierarchy level, and a 7.8% increase in F1-score for drug perturbation classification.

Abstract

Hierarchical structure is common in image data, where fine-grained clusters often merge into larger, coarser semantic groups. In biological cell images, current self-supervised learning models often suppress this hierarchy, as coarse factors such as imaging modality can obscure finer morphological attributes in the latent space. We propose a hierarchy-aware self-supervised training framework to address this problem. Our method combines two components: a distillation framework with a segmentation teacher to improve morphological awareness in the latent space, and a hierarchy-aware contrastive loss based on HDBSCAN to improve decision boundaries between closely related subtypes at different hierarchical levels. Together, these components reduce the tendency of self-supervised learning to overemphasize coarse factors and instead align embeddings with semantic and morphological cues. This yields biologically meaningful sub-clusters driven by fine morphological detail. We train and evaluate our method on a curated corpus of 2.3 million single cells aggregated from 20 microscopy datasets, both labeled and unlabeled, covering 208 cell classes. Our method improves over baseline and counterpart methods, increasing average top-K accuracy by 2.8%, top-9 retrieval on the dataset with the deepest hierarchy by 6.3%, and downstream F1-score for biologically relevant drug classification from perturbed cell morphology by 7.8%.

Problem

• hierarchical structure
• self-supervised learning
• single cell microscopy
• morphological attributes
• latent space

Innovation

• hierarchy-aware
• self-supervised learning
• single-cell microscopy
• contrastive loss
• morphological embedding